When a user carries out on-line shopping, he/she typically starts from searching for products on e-commerce websites. A user inputs certain keywords corresponding to his/her desires and the returned search result would typically include recommended categories of products and a list of products. In particular, the categories of products may include front-end categories and back-end categories. The front-end categories are for user interface (UI) presentation, while the back-end categories are for management products. Currently, categorization of products in mainstream systems is generally presented in a tree-like structure in which each parent-category has a plurality of sub-categories, while each sub-category has only one parent-category. Moreover, the scopes of the categories decrease from top of the tree towards the bottom of the tree.
In the e-commerce websites of earlier days, the recommended categories were determined according to the number of products returned in the search result under the keyword used by the user and were typically shown in hierarchies. However, with the rapid increase in the number of products, when a user inputs certain keywords, the number of categories has increased dramatically and as a result the user may have difficulty in deciding under which category or categories he/she should conduct a refined search. To solve the above problem, one approach is to score the relevance of categories according to past history of user behavior of clicking on the various categories and to present the categories dynamically based on the scores. The relevance of categories typically decreases from left to right and the categories with less relevant may be hidden by category folding. However, the above approach still presents the search results starting from the first-level categories, and consequently the user needs to click many times to screen the results into liner categories. For example, when a user inputs the keyword “T-shirt”, the search result may present first-level categories such as “women's wear”, “men's wear” and “others.” As such, the user needs to click on one of the first-level categories, e.g., “women's wear”, in order to view refined categories such as “short-sleeve T-shirt,” “couple's T-shirt,” “cotton T-shirt,” etc.
In order to shorten search time, most e-commerce websites nowadays use intelligent navigation techniques to facilitate a user search. Intelligent navigation employs a bottom-up approach for recommendation, which considers all the factors such as number of clicks, purchases of products of a certain category, and number of products corresponding to the keywords, and provides the categories or features which are most relevant according to a certain recommendation algorithm. When the number of clicks, purchases or products of a certain category, or number of products reaches a gives threshold, the bottom-up process is stopped. However, certain drawbacks still exist in the above approach which obtains the recommended categories by calculating data relating to the user. The drawbacks include factors like: noise interference in the data of user behavior and misplacement of one or more categories affecting the accuracy of the recommended categories; the recommended categories being not rich enough when the data of user behavior related to the keyword is small; and the inability to provide recommendation when the number of clicks corresponding to the keyword is lacking.
To solve the aforementioned drawbacks, artificial factors can be added into the recommended categories by intelligent navigation when improving the recommendation algorithm. Conventional techniques mainly implement artificial interference of intelligent navigation by editing the recommended terms. Website operators write the keywords that need to be edited and the recommended categories into a text document in a pre-defined format, integrate the artificial data with the data recommended by the algorithm, and save the integrated data in a server used for the intelligent navigation.
The following problems, however, still exit in the conventional techniques: the artificial interference of intelligent navigation lacks a user-friendly interface; propensity of errors in recommendations due to the recommendations being made based on the experience of website operators without data support; and the artificial interference of intelligent navigation lacking feedback from users and hence the inability to trace the effect of the artificial interference.